Using deep learning to predict tyrosinase inhibitors for treating hyperpigmentation disorders

Background In recent years, skin health has garnered widespread attention, with hyperpigmentation disorders caused by excessive melanin deposition emerging as a particularly prominent concern. Tyrosinase, as the rate-limiting enzyme in the melanin synthesis process, has long been a major focus in the development of its inhibitors. However, only a limited number of tyrosinase inhibitors are currently available for clinical treatment of such disorders, and they are associated with certain toxicity concerns. Consequently, there is an urgent need to develop novel inhibitors that combine high efficacy with low toxicity. Recent studies have shown that deep learning technology exhibits strong capabilities in uncovering the intrinsic patterns of data and predicting the biological activities of compounds, providing a significant opportunity for the rapid screening of novel tyrosinase inhibitors. Methods Based on a dataset of tyrosinase-related compounds, this study constructed a deep learning model to predict compounds that inhibit tyrosinase activity. Using this model, we conducted activity predictions for 36,585 compound and selected the top 100 molecules with the highest prediction scores for screening and verification. Results Literature comparison revealed that 53 of these molecules had been reported to inhibit tyrosinase activity, providing initial support for the model’s reliability. After further screening based on specific criteria, 10 candidate molecules were ultimately selected for molecular docking studies. The docking results indicated that these molecules had good binding potential with the target protein, indirectly supporting the accuracy of the model’s prediction. The final experimental verification revealed that compounds 5 and 10 significantly inhibited tyrosinase activity and reduced melanin content.